Mistral Large 4 opens its weights, with a condition for model services
Le Chonk can now be deployed on your infrastructure. The announcement separates open weights, API use and a commercial licence condition for Model-as-a-Service businesses.
Going deeper
Generated image · SecuFocus
At a glance
Key points
On 29 October, Mistral announced downloadable improved weights for Mistral Large 4, nicknamed Le Chonk. Running it on your infrastructure is now an alternative to sending requests to its API. The announcement does not give every business the same rights: companies operating a Model-as-a-Service offering need a separate commercial licence before commercial use.
Open weights change where inference can run
Mistral distinguishes the public preview announced earlier this month from the improved weights now released. Its announcement points to weight downloads and deployment with vLLM or SGLang. It describes a multimodal model with one trillion parameters. That number alone does not establish a RAM requirement or how many GPUs to buy.
The choice is practical: API requests go to the selected service, while operating the weights lets you choose the infrastructure receiving them. You also take responsibility for operation, access controls and data. We downloaded no weights and measured no server deployment.
Open weights do not give every business identical rights
The announcement names the Mistral Ecosystem Open-Weight license, which the publisher describes as a modified MIT licence. It says use, modification, fine-tuning and commercialization are allowed, with one condition: businesses operating a Model-as-a-Service offering must obtain a separate commercial licence before using the model commercially.
Do not read this as an ordinary MIT licence with no additional condition. If you operate a model service, have the full licence and your intended activity checked before commercial deployment. This brief reports the published condition; it is not a contractual assessment of your project.
The API remains a separate, token-priced option
The announcement lists API prices of USD0.70 per million input tokens, USD0.07 for cached input and USD2.10 for output. These are Mistral’s announced rates, not a checked invoice. They do not price running the weights on your own infrastructure.
Mistral also publishes coding and security results. We do not present them as independent tests or a guarantee against agent hijacking. Keep tool controls in place when changing models and evaluate your own tasks with suitable data.
Check and explore
Sources for this article
Numbers connect each reference to the passages that use it. Dates show when the documentation was consulted.
This article draws on the sources above. The exercises are for you to try on your devices; SecuFocus does not present them as tests carried out by its editorial team. Interfaces and features can change. Method and corrections.
Cite this article
Keep this reference with the article when you save or share it.